@Lisandro Zampella 🤦‍♂️

@Lisandro Zampella cool

I worked in the industry for many years and here are the key things you need to know. US: AI voice = illegal w/o written consent (TCPA). Live AI-assisted = gray area but safer if no synthetic voice + human-triggered. B2B calls have more leeway but must honor DNC + give opt-out. CA: AI/prerecorded = banned w/o consent. Live cold calls (esp. B2B) are fine if you ID yourself + respect opt-out/DNC. UK/EU: AI voice to consumers = no. B2B is allowed under “legit interest” if you ID yourself, allow opt-out, avoid TPS numbers, and disclose AI use if it impacts the interaction. Bottom line: consent rules vary, but no AI voice w/o consent anywhere. Always give opt-out, always log DNCs, and don’t use shady lists.

I have done something similar for a travel agency project. I think I know what you mean. And is it embedded into a website or what type of integration is it? It would be essentially be input to agent, agent trigger tools/other agents(booking management, emails), output back respond (webhook, embedded chat, etc.). If you have the budget you can then combine it with Eleven Lab’s live conversational agent for live voice calls.

@Lisandro Zampella I did it myself, not aware if there is any real videos for it since people gatekeep that knowledge. But if you have enough n8n basic knowledge it’s completely feasible, you can def try with what I said, eleven labs webhook to voice agent, agentic ai for tool control, output back to webhook for voice agent. Simple, but live voice agents are costly if not enterprise, and you have to be clear on prompting and may require multiple agents to accomplish complex tasks, Have fun.

Hey Jeremy — yep, you can totally build this kind of RAG system in n8n without using Google Docs. The full flow would look something like: 1. User uploads a book (PDF, EPUB, DOCX, etc.) 2. Preprocess and convert it to plain text (you’ll need file type detection + parsing) 3. Chunk the text using a recursive splitter 4. Send the chunks to OpenAI’s Embedding API 5. Store the embeddings in Pinecone, with metadata linking to the user 6. When user asks a question, embed the query, search Pinecone, pull relevant chunks, and send those to GPT-4 for the final answer n8n has built-in nodes for OpenAI and Pinecone so it’s fully doable. Supabase works great as a backend for storing transcripts, user auth, etc.

@Christiana Adorable btw what you are doing is against group rules.

@Alejandro Gidi yeah in that case you can still run what I suggested and skip the first part and just straight download the file and run them through the transcriber to the agent.